On feature selection and blast furnace temperature tendency prediction in hot metal based on SVM-RFE

With datasets from the domestic medium blast furnace (BF) as a sample place, the contributions of multivariable features of BF system to temperature tendency prediction are analyzed based on the support vector machine and recursive feature elimination (SVM-RFE), and then prediction model of BF temperature is built. First, the initial feature sets are trained to obtain the optimal feature nested subset based on SVM-RFE. Then, the optimal feature nested subset and the current BF temperature tendency are taken as input and output respectively to build support vector machine (SVM) model, which is applied to the independent test set. Third, the optimal feature set and tendency prediction rate are obtained. The simulation results show that the complexity of high dimension data is reduced. In addition, the model can provide an accuracy of 86% in temperature tendency prediction in BF and have some practical use in online monitoring the BF temperature, and thus it has remarkable advantages in feature selection and BF temperature tendency prediction in hot metal.